Key Parameter Of Injector Powered Transonic Wind Tunnel Modeling

Wind tunnel is a device in aerodynamic experiments of aircraft designed in the preliminary stage. During the high-speed aerodynamic experiment, the stability and precision of wind tunnel’s key parameters such as Mach number, total pressure, static pressure and so on play an important role in the experimental results. So building a highly accurate wind tunnel model has important significance in the flow control of wind tunnel system.In this paper, on the basis of the analysis of ejector transonic wind tunnel aerodynamic structure system, air circulation, and features of experiments, we can find the main factors of the wind tunnel system. Taking into account the three-dimensional air flow, the complex working mechanism and multiple factors, it is difficult to establish the mechanism model. So we select the modeling methods based on data. In this paper, we choose the neural network as the model of key parameters.Through the analysis of the wind tunnel air circulation and wind tunnel experiments, the wind tunnel system is a nonlinear dynamic system, in order to describe the dynamic characteristics of the system effectively, we select Elman neural network, which has state feedback, as the system identification model. The structure parameters of neural network have an important influence on the identification results of network. Structural parameters include the number of hidden layer nodes and the number of input layer nodes. The number of hidden layer node can be determined by several tests, and the number of input layer nodes relevant with the input variables and their orders. The input variables can be determined through analysis of influence factors and the orders can be determined by reconstructing the phase space. In the paper we the obtain variables’ sample time first using mutual information algorithm. Then using Cao algorithm determines the orders of variables.The training algorithm Elman recurrent neural network is based on gradient decline principle which has high computational complexity and slow speed of convergence. A new recursive neural network named echo state network can solve the problem that Elman neural network is difficult to train to some extent. Echo state network only need training the output layer weights and its algorithm is simple which can simplifies the process of training greatly. In order to improve the accuracy of the model and shorten the training time, we use the echo state network as the system identification model for modeling and simulation. Though comparing the simulation results, we finally choose of echo state network as the model of the key parameters of wind tunnel system.